Interesting difference on my hardware between operating systems.

Message boards : Number crunching : Interesting difference on my hardware between operating systems.

To post messages, you must log in.

AuthorMessage
Todd Madson

Send message
Joined: 4 Aug 99
Posts: 63
Credit: 25,150,964
RAC: 4,894
United States
Message 1272593 - Posted: 18 Aug 2012, 16:27:58 UTC

I have played around a bit with running Seti on my i7 based machine in both MacOS and Windows and experienced very interesting differences in terms of what the benchmarker found:

Hardware: 2.8 ghz Intel Corei7, 16 megs PC1333 ram, ATI Radeon HD 4850 with 512 megs video memory.

Windows stats:
Memory 16373.9 MB
Cache 256 KB
Measured floating point speed 2675.11 million ops/sec
Measured integer speed 10577.69 million ops/sec

MacOS stats:
Memory 16384 MB
Cache 976.56 KB
Measured floating point speed 2968.24 million ops/sec
Measured integer speed 4781.98 million ops/sec

In looking at this purely black and white it implies windows mode is going to give me higher performance due to much more efficient integer speed but also on the Windows side I can use my Radeon 4850 to crunch milkyway @ home blocks (I can't seemingly crunch any seti units as the card may be too old to do anything on the seti side due to lack of OpenCL features).

I also on the Mac side can't crunch anything using the video card due to lack of development for that feature on that platform.

Thoughts?


ID: 1272593 · Report as offensive
ClaggyProject Donor
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4622
Credit: 46,334,404
RAC: 2,806
United Kingdom
Message 1272624 - Posted: 18 Aug 2012, 17:38:56 UTC - in response to Message 1272593.

In looking at this purely black and white it implies windows mode is going to give me higher performance due to much more efficient integer speed but also on the Windows side I can use my Radeon 4850 to crunch milkyway @ home blocks (I can't seemingly crunch any seti units as the card may be too old to do anything on the seti side due to lack of OpenCL features).

If you enable Astropulse_v6 work fetch in your preferences you should be able to run the new Stock ATI OpenCL Astropulse app,

Claggy

ID: 1272624 · Report as offensive
Profile arkaynProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4097
Credit: 51,576,090
RAC: 1,593
United States
Message 1272649 - Posted: 18 Aug 2012, 19:03:29 UTC - in response to Message 1272624.

In looking at this purely black and white it implies windows mode is going to give me higher performance due to much more efficient integer speed but also on the Windows side I can use my Radeon 4850 to crunch milkyway @ home blocks (I can't seemingly crunch any seti units as the card may be too old to do anything on the seti side due to lack of OpenCL features).

If you enable Astropulse_v6 work fetch in your preferences you should be able to run the new Stock ATI OpenCL Astropulse app,

Claggy


On the Windows side for now.

Eric is trying to build a Mac version at the moment.

ID: 1272649 · Report as offensive
Richard HaselgroveProject Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 11136
Credit: 83,510,937
RAC: 41,350
United Kingdom
Message 1272716 - Posted: 18 Aug 2012, 20:29:17 UTC - in response to Message 1272593.

In looking at this purely black and white it implies windows mode is going to give me higher performance due to much more efficient integer speed but also on the Windows side I can use my Radeon 4850 to crunch milkyway @ home blocks (I can't seemingly crunch any seti units as the card may be too old to do anything on the seti side due to lack of OpenCL features).

I also on the Mac side can't crunch anything using the video card due to lack of development for that feature on that platform.

Thoughts?

Or it implies that BOINC's benchmarking code is less than 100% consistent cross-platform.

ID: 1272716 · Report as offensive
Ianab
Volunteer tester

Send message
Joined: 11 Jun 08
Posts: 690
Credit: 16,339,829
RAC: 8,809
New Zealand
Message 1272762 - Posted: 18 Aug 2012, 23:44:38 UTC - in response to Message 1272716.

Or it implies that BOINC's benchmarking code is more like a random number generator....

ID: 1272762 · Report as offensive
Profile Ex
Volunteer tester
Avatar

Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 2,077,912
RAC: 435
United States
Message 1272804 - Posted: 19 Aug 2012, 4:08:32 UTC

Yes I've been told by others that the benchmarking is more or less meaningless. the only way you can compare across platforms is to run both platforms for a period of time, and compare the cumulative results to one another.


-Dave #2

3.2.0-33

ID: 1272804 · Report as offensive
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1790
Credit: 225,266,382
RAC: 10,172
Australia
Message 1272827 - Posted: 19 Aug 2012, 6:06:31 UTC

In Windows vs Linux testing I've done using the respective AKv8 CPU apps, given the same hardware I would say the difference in crunching time between any 2 OS's would only be a few percent.

T.A.

ID: 1272827 · Report as offensive
Profile ML1
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 9201
Credit: 5,927,135
RAC: 1,919
United Kingdom
Message 1273402 - Posted: 20 Aug 2012, 17:30:06 UTC

Don't be mislead by the Boinc 'measurements'... The Boinc benchmarks and credits have been a long long story of controversy and confusion...


Happy fast crunchin',
Martin


See new freedom: Mageia5
See & try out for yourself: Linux Voice
The Future is what We all make IT (GPLv3)

ID: 1273402 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6085
Credit: 154,987,643
RAC: 47,317
United States
Message 1273479 - Posted: 20 Aug 2012, 19:46:50 UTC

I let BOINC run it's initial benchmark. Then I set the values to what they should be in the client_state.xml. After that I run BOINC with the skip cpu benchmarks command.
I found in testing that different versions of BOINC will generate drastically different numbers. I think they use the same random number generator as they do in the credit granted code.


SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

ID: 1273479 · Report as offensive
WinterKnight
Volunteer tester

Send message
Joined: 18 May 99
Posts: 10168
Credit: 30,526,968
RAC: 3,785
United Kingdom
Message 1273482 - Posted: 20 Aug 2012, 19:51:37 UTC

The differences in benchmark numbers has been under discusion since, at least, msg 1819 25 Jun 2004 | 11:13:14 UTC.

ID: 1273482 · Report as offensive
Todd Madson

Send message
Joined: 4 Aug 99
Posts: 63
Credit: 25,150,964
RAC: 4,894
United States
Message 1273967 - Posted: 22 Aug 2012, 14:43:44 UTC

Given that the windows side is the only side that currently allows video card crunching you would think it would actually provide better overall statistics.

Thinking, thinking...


ID: 1273967 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3654
Credit: 8,590,905
RAC: 1,273
Bulgaria
Message 1274203 - Posted: 23 Aug 2012, 3:40:23 UTC - in response to Message 1273967.
Last modified: 23 Aug 2012, 4:04:09 UTC

Given that the windows side is the only side that currently allows video card crunching you would think it would actually provide better overall statistics.

Thinking, thinking...


Thinking??
What "overall statistics"?

Benchmarks are done by boinc.exe
They are done only on the CPU
BOINC do not know how to compute on GPU (= no code in boinc.exe to do GPU Benchmarks)

And any external process (program) that is using the CPU at the time of Benchmarks affects the values.

E.g. start compressing some big file/folder (WinRAR, 7-Zip, ...) (since you have many CPU cores you may need to start several WinRAR, I'm not sure how many cores it uses for a single process),
[you may also just start WinRAR, press Alt+B (= WinRAR Benchmarks), select 'Multithreading']
[in 7-Zip it is Tools -> Benchmark, select 'Number of CPU threads:' to match your CPU]

Run the BOINC Benchmarks (from menu),
See the values.





- ALF - "Find out what you don't do well ..... then don't do it!" :)

ID: 1274203 · Report as offensive
Todd Madson

Send message
Joined: 4 Aug 99
Posts: 63
Credit: 25,150,964
RAC: 4,894
United States
Message 1274554 - Posted: 23 Aug 2012, 20:29:54 UTC

Statistics was probably the wrong work for it.

I meant "more overall work produced".

I was referring to the fact that if I run my system under windows I get:

-8 Seti @ Home work units being crunched simultaneously plus
-1 Milkyway @ Home work unit crunched.


Versus MacOS side:

-8 Seti @ Home work units being crunched simultaneously. Period.

Mainly because to my knowledge there is no Boinc add-ons at this time to allow
video card crunching on the Mac side. Yet.

I haven't tried astropulse v6 workunits yet but it crunches a workunit for milkyway in 4-7 minutes(!) per workunit versus a couple hours on the actual CPU or on the i7 laptop I have that has a nvidia video card in it (and the cuda
algorithm isn't as fast as that ATI card even though it's faster than the cpu
as well).


ID: 1274554 · Report as offensive
Profile arkaynProject Donor
Volunteer tester
Avatar

Send message
Joined: 14 May 99
Posts: 4097
Credit: 51,576,090
RAC: 1,593
United States
Message 1274581 - Posted: 23 Aug 2012, 21:30:21 UTC - in response to Message 1274554.

I know Collatz has a OpenCL app for OSX as well as a Cuda app.



ID: 1274581 · Report as offensive

Message boards : Number crunching : Interesting difference on my hardware between operating systems.


 
©2016 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.